How can we Make CI360 DM Agent Highly Available

[Shubham5497]

Hi 

My Requirement is to make the CI360 DM agent highly available so if one agent goes down then the other DM agent should do the task. We can make the SAS services highly available through Grid Manager or clustering but don't Know how we can make the DM agent highly available. 

Is it possible to make CI360 Marketing agents Highly available, If yes ,then How?

Thanks for your response in advance

[jcawesome]

I'm also interested to know how this can be done 🙂

[JamesAnderson]

Hi @Shubham5497 ,

In the case of the DM agent, Linux will do a good job via systemd to restart a failed instance. If there is something else going wrong with the DM Agent (for example unable to connect to CI360 cloud services or local SAS 9 install) having another instance of it is unlikely to provide any additional fault tolerance as the problem is external to the agent. In the context of batch execution of campaigns (that is measured in many minutes), waiting a few minutes for the operating system to retry a failed service is reasonable. Note that a Task/Segment that is executing when a DM agent fails is not going to failover to another instance. The failure needs manual intervention to resolve (i.e user resubmits it).

If you start more than one instance of DM Agent (first ensuring you have configured them to run on different ports and with a diffferent instanceid via the cionprem.properties file) they will all successfully connect to the tenant and participate in Task and Segment Map processing. Workload will be shared amongst the agent (not load balanced but each will process requests based on reading the next one on the queue on a first come first served basis). There is no guarantee of all tasks for a specific user session being processed by a single agent.

Importantly the Metadata generation process (dm -gmd) is not supported when running multiple DM Agents. So if you need to do this, you must ensure you only have a single DM Agent running until it completes. 

There is not any documentation on this, so official support may be hard to find.

Regards

James

[jcawesome]

Hi James,

I would also assume that in order to make it 'highly available', you'll need to have separate instances of the agent placed in different midtier servers? if all placed in the same server, then it defeats the purpose.

"Importantly the Metadata generation process (dm -gmd) is not supported when running multiple DM Agents."
If the dm.sh -gmd is run in a midtier server with only one instance of the agent (given that the other instance of the agent is in a separate server), should this still be a problem?

Regards,
JC

[JamesAnderson]

Hi @jcawesome ,

Having the DM agent on a different machine will increase fault tolerance when you lose a machine yes, but one instance per machine doesn not solve the dm -gmd support limitation. Only a single DM Agent instance can be connected to the tenant when running the metadata generation process.

Cheers

James